2 * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 #include <linux/module.h>
33 #include <linux/moduleparam.h>
34 #include <linux/debugfs.h>
35 #include <linux/vmalloc.h>
36 #include <linux/math64.h>
38 #include <rdma/ib_verbs.h>
42 #define DRV_VERSION "0.1"
44 MODULE_AUTHOR("Steve Wise");
45 MODULE_DESCRIPTION("Chelsio T4/T5 RDMA Driver");
46 MODULE_LICENSE("Dual BSD/GPL");
47 MODULE_VERSION(DRV_VERSION);
49 static int allow_db_fc_on_t5;
50 module_param(allow_db_fc_on_t5, int, 0644);
51 MODULE_PARM_DESC(allow_db_fc_on_t5,
52 "Allow DB Flow Control on T5 (default = 0)");
54 static int allow_db_coalescing_on_t5;
55 module_param(allow_db_coalescing_on_t5, int, 0644);
56 MODULE_PARM_DESC(allow_db_coalescing_on_t5,
57 "Allow DB Coalescing on T5 (default = 0)");
60 module_param(c4iw_wr_log, int, 0444);
61 MODULE_PARM_DESC(c4iw_wr_log, "Enables logging of work request timing data.");
63 static int c4iw_wr_log_size_order = 12;
64 module_param(c4iw_wr_log_size_order, int, 0444);
65 MODULE_PARM_DESC(c4iw_wr_log_size_order,
66 "Number of entries (log2) in the work request timing log.");
69 struct list_head entry;
70 struct cxgb4_lld_info lldi;
74 static LIST_HEAD(uld_ctx_list);
75 static DEFINE_MUTEX(dev_mutex);
77 #define DB_FC_RESUME_SIZE 64
78 #define DB_FC_RESUME_DELAY 1
79 #define DB_FC_DRAIN_THRESH 0
81 static struct dentry *c4iw_debugfs_root;
83 struct c4iw_debugfs_data {
84 struct c4iw_dev *devp;
90 static int count_idrs(int id, void *p, void *data)
94 *countp = *countp + 1;
98 static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
101 struct c4iw_debugfs_data *d = file->private_data;
103 return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
106 void c4iw_log_wr_stats(struct t4_wq *wq, struct t4_cqe *cqe)
108 struct wr_log_entry le;
111 if (!wq->rdev->wr_log)
114 idx = (atomic_inc_return(&wq->rdev->wr_log_idx) - 1) &
115 (wq->rdev->wr_log_size - 1);
116 le.poll_sge_ts = cxgb4_read_sge_timestamp(wq->rdev->lldi.ports[0]);
117 getnstimeofday(&le.poll_host_ts);
119 le.cqe_sge_ts = CQE_TS(cqe);
122 le.opcode = CQE_OPCODE(cqe);
123 le.post_host_ts = wq->sq.sw_sq[wq->sq.cidx].host_ts;
124 le.post_sge_ts = wq->sq.sw_sq[wq->sq.cidx].sge_ts;
125 le.wr_id = CQE_WRID_SQ_IDX(cqe);
128 le.opcode = FW_RI_RECEIVE;
129 le.post_host_ts = wq->rq.sw_rq[wq->rq.cidx].host_ts;
130 le.post_sge_ts = wq->rq.sw_rq[wq->rq.cidx].sge_ts;
131 le.wr_id = CQE_WRID_MSN(cqe);
133 wq->rdev->wr_log[idx] = le;
136 static int wr_log_show(struct seq_file *seq, void *v)
138 struct c4iw_dev *dev = seq->private;
139 struct timespec prev_ts = {0, 0};
140 struct wr_log_entry *lep;
144 #define ts2ns(ts) div64_u64((ts) * dev->rdev.lldi.cclk_ps, 1000)
146 idx = atomic_read(&dev->rdev.wr_log_idx) &
147 (dev->rdev.wr_log_size - 1);
150 end = dev->rdev.wr_log_size - 1;
151 lep = &dev->rdev.wr_log[idx];
156 prev_ts = lep->poll_host_ts;
158 seq_printf(seq, "%04u: sec %lu nsec %lu qid %u opcode "
159 "%u %s 0x%x host_wr_delta sec %lu nsec %lu "
160 "post_sge_ts 0x%llx cqe_sge_ts 0x%llx "
161 "poll_sge_ts 0x%llx post_poll_delta_ns %llu "
162 "cqe_poll_delta_ns %llu\n",
164 timespec_sub(lep->poll_host_ts,
166 timespec_sub(lep->poll_host_ts,
168 lep->qid, lep->opcode,
169 lep->opcode == FW_RI_RECEIVE ?
172 timespec_sub(lep->poll_host_ts,
173 lep->post_host_ts).tv_sec,
174 timespec_sub(lep->poll_host_ts,
175 lep->post_host_ts).tv_nsec,
176 lep->post_sge_ts, lep->cqe_sge_ts,
178 ts2ns(lep->poll_sge_ts - lep->post_sge_ts),
179 ts2ns(lep->poll_sge_ts - lep->cqe_sge_ts));
180 prev_ts = lep->poll_host_ts;
183 if (idx > (dev->rdev.wr_log_size - 1))
185 lep = &dev->rdev.wr_log[idx];
191 static int wr_log_open(struct inode *inode, struct file *file)
193 return single_open(file, wr_log_show, inode->i_private);
196 static ssize_t wr_log_clear(struct file *file, const char __user *buf,
197 size_t count, loff_t *pos)
199 struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
202 if (dev->rdev.wr_log)
203 for (i = 0; i < dev->rdev.wr_log_size; i++)
204 dev->rdev.wr_log[i].valid = 0;
208 static const struct file_operations wr_log_debugfs_fops = {
209 .owner = THIS_MODULE,
211 .release = single_release,
214 .write = wr_log_clear,
217 static struct sockaddr_in zero_sin = {
218 .sin_family = AF_INET,
221 static struct sockaddr_in6 zero_sin6 = {
222 .sin6_family = AF_INET6,
225 static void set_ep_sin_addrs(struct c4iw_ep *ep,
226 struct sockaddr_in **lsin,
227 struct sockaddr_in **rsin,
228 struct sockaddr_in **m_lsin,
229 struct sockaddr_in **m_rsin)
231 struct iw_cm_id *id = ep->com.cm_id;
233 *lsin = (struct sockaddr_in *)&ep->com.local_addr;
234 *rsin = (struct sockaddr_in *)&ep->com.remote_addr;
236 *m_lsin = (struct sockaddr_in *)&id->m_local_addr;
237 *m_rsin = (struct sockaddr_in *)&id->m_remote_addr;
244 static void set_ep_sin6_addrs(struct c4iw_ep *ep,
245 struct sockaddr_in6 **lsin6,
246 struct sockaddr_in6 **rsin6,
247 struct sockaddr_in6 **m_lsin6,
248 struct sockaddr_in6 **m_rsin6)
250 struct iw_cm_id *id = ep->com.cm_id;
252 *lsin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
253 *rsin6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
255 *m_lsin6 = (struct sockaddr_in6 *)&id->m_local_addr;
256 *m_rsin6 = (struct sockaddr_in6 *)&id->m_remote_addr;
258 *m_lsin6 = &zero_sin6;
259 *m_rsin6 = &zero_sin6;
263 static int dump_qp(int id, void *p, void *data)
265 struct c4iw_qp *qp = p;
266 struct c4iw_debugfs_data *qpd = data;
270 if (id != qp->wq.sq.qid)
273 space = qpd->bufsize - qpd->pos - 1;
278 struct c4iw_ep *ep = qp->ep;
280 if (ep->com.local_addr.ss_family == AF_INET) {
281 struct sockaddr_in *lsin;
282 struct sockaddr_in *rsin;
283 struct sockaddr_in *m_lsin;
284 struct sockaddr_in *m_rsin;
286 set_ep_sin_addrs(ep, &lsin, &rsin, &m_lsin, &m_rsin);
287 cc = snprintf(qpd->buf + qpd->pos, space,
288 "rc qp sq id %u rq id %u state %u "
289 "onchip %u ep tid %u state %u "
290 "%pI4:%u/%u->%pI4:%u/%u\n",
291 qp->wq.sq.qid, qp->wq.rq.qid,
293 qp->wq.sq.flags & T4_SQ_ONCHIP,
294 ep->hwtid, (int)ep->com.state,
295 &lsin->sin_addr, ntohs(lsin->sin_port),
296 ntohs(m_lsin->sin_port),
297 &rsin->sin_addr, ntohs(rsin->sin_port),
298 ntohs(m_rsin->sin_port));
300 struct sockaddr_in6 *lsin6;
301 struct sockaddr_in6 *rsin6;
302 struct sockaddr_in6 *m_lsin6;
303 struct sockaddr_in6 *m_rsin6;
305 set_ep_sin6_addrs(ep, &lsin6, &rsin6, &m_lsin6,
307 cc = snprintf(qpd->buf + qpd->pos, space,
308 "rc qp sq id %u rq id %u state %u "
309 "onchip %u ep tid %u state %u "
310 "%pI6:%u/%u->%pI6:%u/%u\n",
311 qp->wq.sq.qid, qp->wq.rq.qid,
313 qp->wq.sq.flags & T4_SQ_ONCHIP,
314 ep->hwtid, (int)ep->com.state,
316 ntohs(lsin6->sin6_port),
317 ntohs(m_lsin6->sin6_port),
319 ntohs(rsin6->sin6_port),
320 ntohs(m_rsin6->sin6_port));
323 cc = snprintf(qpd->buf + qpd->pos, space,
324 "qp sq id %u rq id %u state %u onchip %u\n",
325 qp->wq.sq.qid, qp->wq.rq.qid,
327 qp->wq.sq.flags & T4_SQ_ONCHIP);
333 static int qp_release(struct inode *inode, struct file *file)
335 struct c4iw_debugfs_data *qpd = file->private_data;
337 pr_info("%s null qpd?\n", __func__);
345 static int qp_open(struct inode *inode, struct file *file)
347 struct c4iw_debugfs_data *qpd;
350 qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
354 qpd->devp = inode->i_private;
357 spin_lock_irq(&qpd->devp->lock);
358 idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
359 spin_unlock_irq(&qpd->devp->lock);
361 qpd->bufsize = count * 180;
362 qpd->buf = vmalloc(qpd->bufsize);
368 spin_lock_irq(&qpd->devp->lock);
369 idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
370 spin_unlock_irq(&qpd->devp->lock);
372 qpd->buf[qpd->pos++] = 0;
373 file->private_data = qpd;
377 static const struct file_operations qp_debugfs_fops = {
378 .owner = THIS_MODULE,
380 .release = qp_release,
381 .read = debugfs_read,
382 .llseek = default_llseek,
385 static int dump_stag(int id, void *p, void *data)
387 struct c4iw_debugfs_data *stagd = data;
390 struct fw_ri_tpte tpte;
393 space = stagd->bufsize - stagd->pos - 1;
397 ret = cxgb4_read_tpte(stagd->devp->rdev.lldi.ports[0], (u32)id<<8,
400 dev_err(&stagd->devp->rdev.lldi.pdev->dev,
401 "%s cxgb4_read_tpte err %d\n", __func__, ret);
404 cc = snprintf(stagd->buf + stagd->pos, space,
405 "stag: idx 0x%x valid %d key 0x%x state %d pdid %d "
406 "perm 0x%x ps %d len 0x%llx va 0x%llx\n",
408 FW_RI_TPTE_VALID_G(ntohl(tpte.valid_to_pdid)),
409 FW_RI_TPTE_STAGKEY_G(ntohl(tpte.valid_to_pdid)),
410 FW_RI_TPTE_STAGSTATE_G(ntohl(tpte.valid_to_pdid)),
411 FW_RI_TPTE_PDID_G(ntohl(tpte.valid_to_pdid)),
412 FW_RI_TPTE_PERM_G(ntohl(tpte.locread_to_qpid)),
413 FW_RI_TPTE_PS_G(ntohl(tpte.locread_to_qpid)),
414 ((u64)ntohl(tpte.len_hi) << 32) | ntohl(tpte.len_lo),
415 ((u64)ntohl(tpte.va_hi) << 32) | ntohl(tpte.va_lo_fbo));
421 static int stag_release(struct inode *inode, struct file *file)
423 struct c4iw_debugfs_data *stagd = file->private_data;
425 pr_info("%s null stagd?\n", __func__);
433 static int stag_open(struct inode *inode, struct file *file)
435 struct c4iw_debugfs_data *stagd;
439 stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
444 stagd->devp = inode->i_private;
447 spin_lock_irq(&stagd->devp->lock);
448 idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
449 spin_unlock_irq(&stagd->devp->lock);
451 stagd->bufsize = count * 256;
452 stagd->buf = vmalloc(stagd->bufsize);
458 spin_lock_irq(&stagd->devp->lock);
459 idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
460 spin_unlock_irq(&stagd->devp->lock);
462 stagd->buf[stagd->pos++] = 0;
463 file->private_data = stagd;
471 static const struct file_operations stag_debugfs_fops = {
472 .owner = THIS_MODULE,
474 .release = stag_release,
475 .read = debugfs_read,
476 .llseek = default_llseek,
479 static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
481 static int stats_show(struct seq_file *seq, void *v)
483 struct c4iw_dev *dev = seq->private;
485 seq_printf(seq, " Object: %10s %10s %10s %10s\n", "Total", "Current",
487 seq_printf(seq, " PDID: %10llu %10llu %10llu %10llu\n",
488 dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
489 dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
490 seq_printf(seq, " QID: %10llu %10llu %10llu %10llu\n",
491 dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
492 dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
493 seq_printf(seq, " TPTMEM: %10llu %10llu %10llu %10llu\n",
494 dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
495 dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
496 seq_printf(seq, " PBLMEM: %10llu %10llu %10llu %10llu\n",
497 dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
498 dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
499 seq_printf(seq, " RQTMEM: %10llu %10llu %10llu %10llu\n",
500 dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
501 dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
502 seq_printf(seq, " OCQPMEM: %10llu %10llu %10llu %10llu\n",
503 dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
504 dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
505 seq_printf(seq, " DB FULL: %10llu\n", dev->rdev.stats.db_full);
506 seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
507 seq_printf(seq, " DB DROP: %10llu\n", dev->rdev.stats.db_drop);
508 seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
509 db_state_str[dev->db_state],
510 dev->rdev.stats.db_state_transitions,
511 dev->rdev.stats.db_fc_interruptions);
512 seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
513 seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
514 dev->rdev.stats.act_ofld_conn_fails);
515 seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
516 dev->rdev.stats.pas_ofld_conn_fails);
517 seq_printf(seq, "NEG_ADV_RCVD: %10llu\n", dev->rdev.stats.neg_adv);
518 seq_printf(seq, "AVAILABLE IRD: %10u\n", dev->avail_ird);
522 static int stats_open(struct inode *inode, struct file *file)
524 return single_open(file, stats_show, inode->i_private);
527 static ssize_t stats_clear(struct file *file, const char __user *buf,
528 size_t count, loff_t *pos)
530 struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
532 mutex_lock(&dev->rdev.stats.lock);
533 dev->rdev.stats.pd.max = 0;
534 dev->rdev.stats.pd.fail = 0;
535 dev->rdev.stats.qid.max = 0;
536 dev->rdev.stats.qid.fail = 0;
537 dev->rdev.stats.stag.max = 0;
538 dev->rdev.stats.stag.fail = 0;
539 dev->rdev.stats.pbl.max = 0;
540 dev->rdev.stats.pbl.fail = 0;
541 dev->rdev.stats.rqt.max = 0;
542 dev->rdev.stats.rqt.fail = 0;
543 dev->rdev.stats.ocqp.max = 0;
544 dev->rdev.stats.ocqp.fail = 0;
545 dev->rdev.stats.db_full = 0;
546 dev->rdev.stats.db_empty = 0;
547 dev->rdev.stats.db_drop = 0;
548 dev->rdev.stats.db_state_transitions = 0;
549 dev->rdev.stats.tcam_full = 0;
550 dev->rdev.stats.act_ofld_conn_fails = 0;
551 dev->rdev.stats.pas_ofld_conn_fails = 0;
552 mutex_unlock(&dev->rdev.stats.lock);
556 static const struct file_operations stats_debugfs_fops = {
557 .owner = THIS_MODULE,
559 .release = single_release,
562 .write = stats_clear,
565 static int dump_ep(int id, void *p, void *data)
567 struct c4iw_ep *ep = p;
568 struct c4iw_debugfs_data *epd = data;
572 space = epd->bufsize - epd->pos - 1;
576 if (ep->com.local_addr.ss_family == AF_INET) {
577 struct sockaddr_in *lsin;
578 struct sockaddr_in *rsin;
579 struct sockaddr_in *m_lsin;
580 struct sockaddr_in *m_rsin;
582 set_ep_sin_addrs(ep, &lsin, &rsin, &m_lsin, &m_rsin);
583 cc = snprintf(epd->buf + epd->pos, space,
584 "ep %p cm_id %p qp %p state %d flags 0x%lx "
585 "history 0x%lx hwtid %d atid %d "
586 "conn_na %u abort_na %u "
587 "%pI4:%d/%d <-> %pI4:%d/%d\n",
588 ep, ep->com.cm_id, ep->com.qp,
589 (int)ep->com.state, ep->com.flags,
590 ep->com.history, ep->hwtid, ep->atid,
591 ep->stats.connect_neg_adv,
592 ep->stats.abort_neg_adv,
593 &lsin->sin_addr, ntohs(lsin->sin_port),
594 ntohs(m_lsin->sin_port),
595 &rsin->sin_addr, ntohs(rsin->sin_port),
596 ntohs(m_rsin->sin_port));
598 struct sockaddr_in6 *lsin6;
599 struct sockaddr_in6 *rsin6;
600 struct sockaddr_in6 *m_lsin6;
601 struct sockaddr_in6 *m_rsin6;
603 set_ep_sin6_addrs(ep, &lsin6, &rsin6, &m_lsin6, &m_rsin6);
604 cc = snprintf(epd->buf + epd->pos, space,
605 "ep %p cm_id %p qp %p state %d flags 0x%lx "
606 "history 0x%lx hwtid %d atid %d "
607 "conn_na %u abort_na %u "
608 "%pI6:%d/%d <-> %pI6:%d/%d\n",
609 ep, ep->com.cm_id, ep->com.qp,
610 (int)ep->com.state, ep->com.flags,
611 ep->com.history, ep->hwtid, ep->atid,
612 ep->stats.connect_neg_adv,
613 ep->stats.abort_neg_adv,
614 &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
615 ntohs(m_lsin6->sin6_port),
616 &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
617 ntohs(m_rsin6->sin6_port));
624 static int dump_listen_ep(int id, void *p, void *data)
626 struct c4iw_listen_ep *ep = p;
627 struct c4iw_debugfs_data *epd = data;
631 space = epd->bufsize - epd->pos - 1;
635 if (ep->com.local_addr.ss_family == AF_INET) {
636 struct sockaddr_in *lsin = (struct sockaddr_in *)
637 &ep->com.cm_id->local_addr;
638 struct sockaddr_in *m_lsin = (struct sockaddr_in *)
639 &ep->com.cm_id->m_local_addr;
641 cc = snprintf(epd->buf + epd->pos, space,
642 "ep %p cm_id %p state %d flags 0x%lx stid %d "
643 "backlog %d %pI4:%d/%d\n",
644 ep, ep->com.cm_id, (int)ep->com.state,
645 ep->com.flags, ep->stid, ep->backlog,
646 &lsin->sin_addr, ntohs(lsin->sin_port),
647 ntohs(m_lsin->sin_port));
649 struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
650 &ep->com.cm_id->local_addr;
651 struct sockaddr_in6 *m_lsin6 = (struct sockaddr_in6 *)
652 &ep->com.cm_id->m_local_addr;
654 cc = snprintf(epd->buf + epd->pos, space,
655 "ep %p cm_id %p state %d flags 0x%lx stid %d "
656 "backlog %d %pI6:%d/%d\n",
657 ep, ep->com.cm_id, (int)ep->com.state,
658 ep->com.flags, ep->stid, ep->backlog,
659 &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
660 ntohs(m_lsin6->sin6_port));
667 static int ep_release(struct inode *inode, struct file *file)
669 struct c4iw_debugfs_data *epd = file->private_data;
671 pr_info("%s null qpd?\n", __func__);
679 static int ep_open(struct inode *inode, struct file *file)
681 struct c4iw_debugfs_data *epd;
685 epd = kmalloc(sizeof(*epd), GFP_KERNEL);
690 epd->devp = inode->i_private;
693 spin_lock_irq(&epd->devp->lock);
694 idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
695 idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
696 idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
697 spin_unlock_irq(&epd->devp->lock);
699 epd->bufsize = count * 240;
700 epd->buf = vmalloc(epd->bufsize);
706 spin_lock_irq(&epd->devp->lock);
707 idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
708 idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
709 idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
710 spin_unlock_irq(&epd->devp->lock);
712 file->private_data = epd;
720 static const struct file_operations ep_debugfs_fops = {
721 .owner = THIS_MODULE,
723 .release = ep_release,
724 .read = debugfs_read,
727 static int setup_debugfs(struct c4iw_dev *devp)
729 if (!devp->debugfs_root)
732 debugfs_create_file_size("qps", S_IWUSR, devp->debugfs_root,
733 (void *)devp, &qp_debugfs_fops, 4096);
735 debugfs_create_file_size("stags", S_IWUSR, devp->debugfs_root,
736 (void *)devp, &stag_debugfs_fops, 4096);
738 debugfs_create_file_size("stats", S_IWUSR, devp->debugfs_root,
739 (void *)devp, &stats_debugfs_fops, 4096);
741 debugfs_create_file_size("eps", S_IWUSR, devp->debugfs_root,
742 (void *)devp, &ep_debugfs_fops, 4096);
745 debugfs_create_file_size("wr_log", S_IWUSR, devp->debugfs_root,
746 (void *)devp, &wr_log_debugfs_fops, 4096);
750 void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
751 struct c4iw_dev_ucontext *uctx)
753 struct list_head *pos, *nxt;
754 struct c4iw_qid_list *entry;
756 mutex_lock(&uctx->lock);
757 list_for_each_safe(pos, nxt, &uctx->qpids) {
758 entry = list_entry(pos, struct c4iw_qid_list, entry);
759 list_del_init(&entry->entry);
760 if (!(entry->qid & rdev->qpmask)) {
761 c4iw_put_resource(&rdev->resource.qid_table,
763 mutex_lock(&rdev->stats.lock);
764 rdev->stats.qid.cur -= rdev->qpmask + 1;
765 mutex_unlock(&rdev->stats.lock);
770 list_for_each_safe(pos, nxt, &uctx->qpids) {
771 entry = list_entry(pos, struct c4iw_qid_list, entry);
772 list_del_init(&entry->entry);
775 mutex_unlock(&uctx->lock);
778 void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
779 struct c4iw_dev_ucontext *uctx)
781 INIT_LIST_HEAD(&uctx->qpids);
782 INIT_LIST_HEAD(&uctx->cqids);
783 mutex_init(&uctx->lock);
786 /* Caller takes care of locking if needed */
787 static int c4iw_rdev_open(struct c4iw_rdev *rdev)
791 c4iw_init_dev_ucontext(rdev, &rdev->uctx);
794 * This implementation assumes udb_density == ucq_density! Eventually
795 * we might need to support this but for now fail the open. Also the
796 * cqid and qpid range must match for now.
798 if (rdev->lldi.udb_density != rdev->lldi.ucq_density) {
799 pr_err("%s: unsupported udb/ucq densities %u/%u\n",
800 pci_name(rdev->lldi.pdev), rdev->lldi.udb_density,
801 rdev->lldi.ucq_density);
804 if (rdev->lldi.vr->qp.start != rdev->lldi.vr->cq.start ||
805 rdev->lldi.vr->qp.size != rdev->lldi.vr->cq.size) {
806 pr_err("%s: unsupported qp and cq id ranges qp start %u size %u cq start %u size %u\n",
807 pci_name(rdev->lldi.pdev), rdev->lldi.vr->qp.start,
808 rdev->lldi.vr->qp.size, rdev->lldi.vr->cq.size,
809 rdev->lldi.vr->cq.size);
813 rdev->qpmask = rdev->lldi.udb_density - 1;
814 rdev->cqmask = rdev->lldi.ucq_density - 1;
815 pr_debug("%s dev %s stag start 0x%0x size 0x%0x num stags %d pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x qp qid start %u size %u cq qid start %u size %u\n",
816 __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
817 rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
818 rdev->lldi.vr->pbl.start,
819 rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
820 rdev->lldi.vr->rq.size,
821 rdev->lldi.vr->qp.start,
822 rdev->lldi.vr->qp.size,
823 rdev->lldi.vr->cq.start,
824 rdev->lldi.vr->cq.size);
825 pr_debug("udb %pR db_reg %p gts_reg %p qpmask 0x%x cqmask 0x%x\n",
826 &rdev->lldi.pdev->resource[2],
827 rdev->lldi.db_reg, rdev->lldi.gts_reg,
828 rdev->qpmask, rdev->cqmask);
830 if (c4iw_num_stags(rdev) == 0)
833 rdev->stats.pd.total = T4_MAX_NUM_PD;
834 rdev->stats.stag.total = rdev->lldi.vr->stag.size;
835 rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
836 rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
837 rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
838 rdev->stats.qid.total = rdev->lldi.vr->qp.size;
840 err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
842 pr_err("error %d initializing resources\n", err);
845 err = c4iw_pblpool_create(rdev);
847 pr_err("error %d initializing pbl pool\n", err);
848 goto destroy_resource;
850 err = c4iw_rqtpool_create(rdev);
852 pr_err("error %d initializing rqt pool\n", err);
853 goto destroy_pblpool;
855 err = c4iw_ocqp_pool_create(rdev);
857 pr_err("error %d initializing ocqp pool\n", err);
858 goto destroy_rqtpool;
860 rdev->status_page = (struct t4_dev_status_page *)
861 __get_free_page(GFP_KERNEL);
862 if (!rdev->status_page) {
864 goto destroy_ocqp_pool;
866 rdev->status_page->qp_start = rdev->lldi.vr->qp.start;
867 rdev->status_page->qp_size = rdev->lldi.vr->qp.size;
868 rdev->status_page->cq_start = rdev->lldi.vr->cq.start;
869 rdev->status_page->cq_size = rdev->lldi.vr->cq.size;
872 rdev->wr_log = kzalloc((1 << c4iw_wr_log_size_order) *
873 sizeof(*rdev->wr_log), GFP_KERNEL);
875 rdev->wr_log_size = 1 << c4iw_wr_log_size_order;
876 atomic_set(&rdev->wr_log_idx, 0);
880 rdev->free_workq = create_singlethread_workqueue("iw_cxgb4_free");
881 if (!rdev->free_workq) {
883 goto err_free_status_page;
886 rdev->status_page->db_off = 0;
889 err_free_status_page:
890 free_page((unsigned long)rdev->status_page);
892 c4iw_ocqp_pool_destroy(rdev);
894 c4iw_rqtpool_destroy(rdev);
896 c4iw_pblpool_destroy(rdev);
898 c4iw_destroy_resource(&rdev->resource);
902 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
904 destroy_workqueue(rdev->free_workq);
906 free_page((unsigned long)rdev->status_page);
907 c4iw_pblpool_destroy(rdev);
908 c4iw_rqtpool_destroy(rdev);
909 c4iw_destroy_resource(&rdev->resource);
912 static void c4iw_dealloc(struct uld_ctx *ctx)
914 c4iw_rdev_close(&ctx->dev->rdev);
915 WARN_ON_ONCE(!idr_is_empty(&ctx->dev->cqidr));
916 idr_destroy(&ctx->dev->cqidr);
917 WARN_ON_ONCE(!idr_is_empty(&ctx->dev->qpidr));
918 idr_destroy(&ctx->dev->qpidr);
919 WARN_ON_ONCE(!idr_is_empty(&ctx->dev->mmidr));
920 idr_destroy(&ctx->dev->mmidr);
921 wait_event(ctx->dev->wait, idr_is_empty(&ctx->dev->hwtid_idr));
922 idr_destroy(&ctx->dev->hwtid_idr);
923 idr_destroy(&ctx->dev->stid_idr);
924 idr_destroy(&ctx->dev->atid_idr);
925 if (ctx->dev->rdev.bar2_kva)
926 iounmap(ctx->dev->rdev.bar2_kva);
927 if (ctx->dev->rdev.oc_mw_kva)
928 iounmap(ctx->dev->rdev.oc_mw_kva);
929 ib_dealloc_device(&ctx->dev->ibdev);
933 static void c4iw_remove(struct uld_ctx *ctx)
935 pr_debug("%s c4iw_dev %p\n", __func__, ctx->dev);
936 c4iw_unregister_device(ctx->dev);
940 static int rdma_supported(const struct cxgb4_lld_info *infop)
942 return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
943 infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
944 infop->vr->cq.size > 0;
947 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
949 struct c4iw_dev *devp;
952 if (!rdma_supported(infop)) {
953 pr_info("%s: RDMA not supported on this device\n",
954 pci_name(infop->pdev));
955 return ERR_PTR(-ENOSYS);
957 if (!ocqp_supported(infop))
958 pr_info("%s: On-Chip Queues not supported on this device\n",
959 pci_name(infop->pdev));
961 devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
963 pr_err("Cannot allocate ib device\n");
964 return ERR_PTR(-ENOMEM);
966 devp->rdev.lldi = *infop;
968 /* init various hw-queue params based on lld info */
969 pr_debug("%s: Ing. padding boundary is %d, egrsstatuspagesize = %d\n",
970 __func__, devp->rdev.lldi.sge_ingpadboundary,
971 devp->rdev.lldi.sge_egrstatuspagesize);
973 devp->rdev.hw_queue.t4_eq_status_entries =
974 devp->rdev.lldi.sge_ingpadboundary > 64 ? 2 : 1;
975 devp->rdev.hw_queue.t4_max_eq_size = 65520;
976 devp->rdev.hw_queue.t4_max_iq_size = 65520;
977 devp->rdev.hw_queue.t4_max_rq_size = 8192 -
978 devp->rdev.hw_queue.t4_eq_status_entries - 1;
979 devp->rdev.hw_queue.t4_max_sq_size =
980 devp->rdev.hw_queue.t4_max_eq_size -
981 devp->rdev.hw_queue.t4_eq_status_entries - 1;
982 devp->rdev.hw_queue.t4_max_qp_depth =
983 devp->rdev.hw_queue.t4_max_rq_size;
984 devp->rdev.hw_queue.t4_max_cq_depth =
985 devp->rdev.hw_queue.t4_max_iq_size - 2;
986 devp->rdev.hw_queue.t4_stat_len =
987 devp->rdev.lldi.sge_egrstatuspagesize;
990 * For T5/T6 devices, we map all of BAR2 with WC.
991 * For T4 devices with onchip qp mem, we map only that part
994 devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
995 if (!is_t4(devp->rdev.lldi.adapter_type)) {
996 devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
997 pci_resource_len(devp->rdev.lldi.pdev, 2));
998 if (!devp->rdev.bar2_kva) {
999 pr_err("Unable to ioremap BAR2\n");
1000 ib_dealloc_device(&devp->ibdev);
1001 return ERR_PTR(-EINVAL);
1003 } else if (ocqp_supported(infop)) {
1004 devp->rdev.oc_mw_pa =
1005 pci_resource_start(devp->rdev.lldi.pdev, 2) +
1006 pci_resource_len(devp->rdev.lldi.pdev, 2) -
1007 roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
1008 devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
1009 devp->rdev.lldi.vr->ocq.size);
1010 if (!devp->rdev.oc_mw_kva) {
1011 pr_err("Unable to ioremap onchip mem\n");
1012 ib_dealloc_device(&devp->ibdev);
1013 return ERR_PTR(-EINVAL);
1017 pr_debug("ocq memory: hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
1018 devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
1019 devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
1021 ret = c4iw_rdev_open(&devp->rdev);
1023 pr_err("Unable to open CXIO rdev err %d\n", ret);
1024 ib_dealloc_device(&devp->ibdev);
1025 return ERR_PTR(ret);
1028 idr_init(&devp->cqidr);
1029 idr_init(&devp->qpidr);
1030 idr_init(&devp->mmidr);
1031 idr_init(&devp->hwtid_idr);
1032 idr_init(&devp->stid_idr);
1033 idr_init(&devp->atid_idr);
1034 spin_lock_init(&devp->lock);
1035 mutex_init(&devp->rdev.stats.lock);
1036 mutex_init(&devp->db_mutex);
1037 INIT_LIST_HEAD(&devp->db_fc_list);
1038 init_waitqueue_head(&devp->wait);
1039 devp->avail_ird = devp->rdev.lldi.max_ird_adapter;
1041 if (c4iw_debugfs_root) {
1042 devp->debugfs_root = debugfs_create_dir(
1043 pci_name(devp->rdev.lldi.pdev),
1045 setup_debugfs(devp);
1052 static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
1054 struct uld_ctx *ctx;
1055 static int vers_printed;
1058 if (!vers_printed++)
1059 pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
1062 ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
1064 ctx = ERR_PTR(-ENOMEM);
1069 pr_debug("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n",
1070 __func__, pci_name(ctx->lldi.pdev),
1071 ctx->lldi.nchan, ctx->lldi.nrxq,
1072 ctx->lldi.ntxq, ctx->lldi.nports);
1074 mutex_lock(&dev_mutex);
1075 list_add_tail(&ctx->entry, &uld_ctx_list);
1076 mutex_unlock(&dev_mutex);
1078 for (i = 0; i < ctx->lldi.nrxq; i++)
1079 pr_debug("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
1084 static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
1088 struct sk_buff *skb;
1091 * Allocate space for cpl_pass_accept_req which will be synthesized by
1092 * driver. Once the driver synthesizes the request the skb will go
1093 * through the regular cpl_pass_accept_req processing.
1094 * The math here assumes sizeof cpl_pass_accept_req >= sizeof
1097 skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1098 sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
1102 __skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
1103 sizeof(struct rss_header) - pktshift);
1106 * This skb will contain:
1107 * rss_header from the rspq descriptor (1 flit)
1108 * cpl_rx_pkt struct from the rspq descriptor (2 flits)
1109 * space for the difference between the size of an
1110 * rx_pkt and pass_accept_req cpl (1 flit)
1111 * the packet data from the gl
1113 skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
1114 sizeof(struct rss_header));
1115 skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
1116 sizeof(struct cpl_pass_accept_req),
1118 gl->tot_len - pktshift);
1122 static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
1125 unsigned int opcode = *(u8 *)rsp;
1126 struct sk_buff *skb;
1128 if (opcode != CPL_RX_PKT)
1131 skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
1135 if (c4iw_handlers[opcode] == NULL) {
1136 pr_info("%s no handler opcode 0x%x...\n", __func__, opcode);
1140 c4iw_handlers[opcode](dev, skb);
1146 static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
1147 const struct pkt_gl *gl)
1149 struct uld_ctx *ctx = handle;
1150 struct c4iw_dev *dev = ctx->dev;
1151 struct sk_buff *skb;
1155 /* omit RSS and rsp_ctrl at end of descriptor */
1156 unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
1158 skb = alloc_skb(256, GFP_ATOMIC);
1161 __skb_put(skb, len);
1162 skb_copy_to_linear_data(skb, &rsp[1], len);
1163 } else if (gl == CXGB4_MSG_AN) {
1164 const struct rsp_ctrl *rc = (void *)rsp;
1166 u32 qid = be32_to_cpu(rc->pldbuflen_qid);
1167 c4iw_ev_handler(dev, qid);
1169 } else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
1170 if (recv_rx_pkt(dev, gl, rsp))
1173 pr_info("%s: unexpected FL contents at %p, RSS %#llx, FL %#llx, len %u\n",
1174 pci_name(ctx->lldi.pdev), gl->va,
1176 be64_to_cpu(*(__force __be64 *)gl->va),
1181 skb = cxgb4_pktgl_to_skb(gl, 128, 128);
1186 opcode = *(u8 *)rsp;
1187 if (c4iw_handlers[opcode]) {
1188 c4iw_handlers[opcode](dev, skb);
1190 pr_info("%s no handler opcode 0x%x...\n", __func__, opcode);
1199 static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
1201 struct uld_ctx *ctx = handle;
1203 pr_debug("%s new_state %u\n", __func__, new_state);
1204 switch (new_state) {
1205 case CXGB4_STATE_UP:
1206 pr_info("%s: Up\n", pci_name(ctx->lldi.pdev));
1210 ctx->dev = c4iw_alloc(&ctx->lldi);
1211 if (IS_ERR(ctx->dev)) {
1212 pr_err("%s: initialization failed: %ld\n",
1213 pci_name(ctx->lldi.pdev),
1218 ret = c4iw_register_device(ctx->dev);
1220 pr_err("%s: RDMA registration failed: %d\n",
1221 pci_name(ctx->lldi.pdev), ret);
1226 case CXGB4_STATE_DOWN:
1227 pr_info("%s: Down\n", pci_name(ctx->lldi.pdev));
1231 case CXGB4_STATE_START_RECOVERY:
1232 pr_info("%s: Fatal Error\n", pci_name(ctx->lldi.pdev));
1234 struct ib_event event;
1236 ctx->dev->rdev.flags |= T4_FATAL_ERROR;
1237 memset(&event, 0, sizeof event);
1238 event.event = IB_EVENT_DEVICE_FATAL;
1239 event.device = &ctx->dev->ibdev;
1240 ib_dispatch_event(&event);
1244 case CXGB4_STATE_DETACH:
1245 pr_info("%s: Detach\n", pci_name(ctx->lldi.pdev));
1253 static int disable_qp_db(int id, void *p, void *data)
1255 struct c4iw_qp *qp = p;
1257 t4_disable_wq_db(&qp->wq);
1261 static void stop_queues(struct uld_ctx *ctx)
1263 unsigned long flags;
1265 spin_lock_irqsave(&ctx->dev->lock, flags);
1266 ctx->dev->rdev.stats.db_state_transitions++;
1267 ctx->dev->db_state = STOPPED;
1268 if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
1269 idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
1271 ctx->dev->rdev.status_page->db_off = 1;
1272 spin_unlock_irqrestore(&ctx->dev->lock, flags);
1275 static int enable_qp_db(int id, void *p, void *data)
1277 struct c4iw_qp *qp = p;
1279 t4_enable_wq_db(&qp->wq);
1283 static void resume_rc_qp(struct c4iw_qp *qp)
1285 spin_lock(&qp->lock);
1286 t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc, NULL);
1287 qp->wq.sq.wq_pidx_inc = 0;
1288 t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc, NULL);
1289 qp->wq.rq.wq_pidx_inc = 0;
1290 spin_unlock(&qp->lock);
1293 static void resume_a_chunk(struct uld_ctx *ctx)
1298 for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
1299 qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
1301 list_del_init(&qp->db_fc_entry);
1303 if (list_empty(&ctx->dev->db_fc_list))
1308 static void resume_queues(struct uld_ctx *ctx)
1310 spin_lock_irq(&ctx->dev->lock);
1311 if (ctx->dev->db_state != STOPPED)
1313 ctx->dev->db_state = FLOW_CONTROL;
1315 if (list_empty(&ctx->dev->db_fc_list)) {
1316 WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
1317 ctx->dev->db_state = NORMAL;
1318 ctx->dev->rdev.stats.db_state_transitions++;
1319 if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
1320 idr_for_each(&ctx->dev->qpidr, enable_qp_db,
1323 ctx->dev->rdev.status_page->db_off = 0;
1327 if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
1328 < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
1329 DB_FC_DRAIN_THRESH)) {
1330 resume_a_chunk(ctx);
1332 if (!list_empty(&ctx->dev->db_fc_list)) {
1333 spin_unlock_irq(&ctx->dev->lock);
1334 if (DB_FC_RESUME_DELAY) {
1335 set_current_state(TASK_UNINTERRUPTIBLE);
1336 schedule_timeout(DB_FC_RESUME_DELAY);
1338 spin_lock_irq(&ctx->dev->lock);
1339 if (ctx->dev->db_state != FLOW_CONTROL)
1345 if (ctx->dev->db_state != NORMAL)
1346 ctx->dev->rdev.stats.db_fc_interruptions++;
1347 spin_unlock_irq(&ctx->dev->lock);
1352 struct c4iw_qp **qps;
1355 static int add_and_ref_qp(int id, void *p, void *data)
1357 struct qp_list *qp_listp = data;
1358 struct c4iw_qp *qp = p;
1360 c4iw_qp_add_ref(&qp->ibqp);
1361 qp_listp->qps[qp_listp->idx++] = qp;
1365 static int count_qps(int id, void *p, void *data)
1367 unsigned *countp = data;
1372 static void deref_qps(struct qp_list *qp_list)
1376 for (idx = 0; idx < qp_list->idx; idx++)
1377 c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
1380 static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
1385 for (idx = 0; idx < qp_list->idx; idx++) {
1386 struct c4iw_qp *qp = qp_list->qps[idx];
1388 spin_lock_irq(&qp->rhp->lock);
1389 spin_lock(&qp->lock);
1390 ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1392 t4_sq_host_wq_pidx(&qp->wq),
1393 t4_sq_wq_size(&qp->wq));
1395 pr_err("%s: Fatal error - DB overflow recovery failed - error syncing SQ qid %u\n",
1396 pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
1397 spin_unlock(&qp->lock);
1398 spin_unlock_irq(&qp->rhp->lock);
1401 qp->wq.sq.wq_pidx_inc = 0;
1403 ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
1405 t4_rq_host_wq_pidx(&qp->wq),
1406 t4_rq_wq_size(&qp->wq));
1409 pr_err("%s: Fatal error - DB overflow recovery failed - error syncing RQ qid %u\n",
1410 pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
1411 spin_unlock(&qp->lock);
1412 spin_unlock_irq(&qp->rhp->lock);
1415 qp->wq.rq.wq_pidx_inc = 0;
1416 spin_unlock(&qp->lock);
1417 spin_unlock_irq(&qp->rhp->lock);
1419 /* Wait for the dbfifo to drain */
1420 while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
1421 set_current_state(TASK_UNINTERRUPTIBLE);
1422 schedule_timeout(usecs_to_jiffies(10));
1427 static void recover_queues(struct uld_ctx *ctx)
1430 struct qp_list qp_list;
1433 /* slow everybody down */
1434 set_current_state(TASK_UNINTERRUPTIBLE);
1435 schedule_timeout(usecs_to_jiffies(1000));
1437 /* flush the SGE contexts */
1438 ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
1440 pr_err("%s: Fatal error - DB overflow recovery failed\n",
1441 pci_name(ctx->lldi.pdev));
1445 /* Count active queues so we can build a list of queues to recover */
1446 spin_lock_irq(&ctx->dev->lock);
1447 WARN_ON(ctx->dev->db_state != STOPPED);
1448 ctx->dev->db_state = RECOVERY;
1449 idr_for_each(&ctx->dev->qpidr, count_qps, &count);
1451 qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
1453 spin_unlock_irq(&ctx->dev->lock);
1458 /* add and ref each qp so it doesn't get freed */
1459 idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
1461 spin_unlock_irq(&ctx->dev->lock);
1463 /* now traverse the list in a safe context to recover the db state*/
1464 recover_lost_dbs(ctx, &qp_list);
1466 /* we're almost done! deref the qps and clean up */
1467 deref_qps(&qp_list);
1470 spin_lock_irq(&ctx->dev->lock);
1471 WARN_ON(ctx->dev->db_state != RECOVERY);
1472 ctx->dev->db_state = STOPPED;
1473 spin_unlock_irq(&ctx->dev->lock);
1476 static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
1478 struct uld_ctx *ctx = handle;
1481 case CXGB4_CONTROL_DB_FULL:
1483 ctx->dev->rdev.stats.db_full++;
1485 case CXGB4_CONTROL_DB_EMPTY:
1487 mutex_lock(&ctx->dev->rdev.stats.lock);
1488 ctx->dev->rdev.stats.db_empty++;
1489 mutex_unlock(&ctx->dev->rdev.stats.lock);
1491 case CXGB4_CONTROL_DB_DROP:
1492 recover_queues(ctx);
1493 mutex_lock(&ctx->dev->rdev.stats.lock);
1494 ctx->dev->rdev.stats.db_drop++;
1495 mutex_unlock(&ctx->dev->rdev.stats.lock);
1498 pr_warn("%s: unknown control cmd %u\n",
1499 pci_name(ctx->lldi.pdev), control);
1505 static struct cxgb4_uld_info c4iw_uld_info = {
1507 .nrxq = MAX_ULD_QSETS,
1508 .ntxq = MAX_ULD_QSETS,
1512 .add = c4iw_uld_add,
1513 .rx_handler = c4iw_uld_rx_handler,
1514 .state_change = c4iw_uld_state_change,
1515 .control = c4iw_uld_control,
1518 static int __init c4iw_init_module(void)
1522 err = c4iw_cm_init();
1526 c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
1527 if (!c4iw_debugfs_root)
1528 pr_warn("could not create debugfs entry, continuing\n");
1530 cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
1535 static void __exit c4iw_exit_module(void)
1537 struct uld_ctx *ctx, *tmp;
1539 mutex_lock(&dev_mutex);
1540 list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
1545 mutex_unlock(&dev_mutex);
1546 cxgb4_unregister_uld(CXGB4_ULD_RDMA);
1548 debugfs_remove_recursive(c4iw_debugfs_root);
1551 module_init(c4iw_init_module);
1552 module_exit(c4iw_exit_module);